TWI679439B - Power management system and method for managing power - Google Patents

Abstract

A power management system and a power management method. The power management system includes a host manager and at least one server. The server communicates with the host manager. The server includes at least one processor, at least one voltage regulator, and a voltage regulator controller. Regulators provide actual power to the corresponding processors. The voltage regulator controller regulates the actual power provided by the voltage regulator. The host manager controls the voltage regulator controller in the server, and uses the voltage regulator controller to adjust the actual power provided by the voltage regulator to manage the power of the processor.

Description

Power management system and power management method

The present invention relates to a power management technology for an electronic device, and in particular, to a power management system and a power management method for managing the power of a processor by using a voltage regulator controller to adjust actual power.

Power management is a very important function for electronic devices (such as personal computers, laptops, servers, etc.) with processors (such as central processing units, image processors, etc.), especially for many devices with many servers. For data centers, the savings that power management can save are even greater. Various processors have their own thermal design power (Thermal Design Power; TDP) and the upper limit of the operating frequency when designing. Taking the central processing unit as an example, in order to perform automatic power management, the central processing unit can use dynamic voltage scaling (DVS) to adjust the operating frequency and operating voltage of the central processing unit according to its own workload when the electronic device is operating. To reduce power consumption.

Current power management can usually only reduce power consumption by reducing the operating frequency and / or power when the CPU's workload is low. When the CPU When the workload is high, because the CPU has an upper power limit and an upper frequency limit, the CPU can only run at full speed based on these upper limits, and there is no other power management approach to solve the problem of higher workload. In other words, in some cases, it may be necessary for the server to increase its workload in a short period of time, but under abnormal circumstances, the current solution can only add additional servers to average the workload, but not Solve it by the existing server temporarily increasing the upper limit of its efficiency.

The invention provides a power management system and a power management method, which can simultaneously set and adjust the actual power of processors in multiple running servers at the same time, so as to instantly increase or limit the overall power consumption of the servers.

The power management system of the present invention includes a host manager and a server. The server is in communication with the host manager. The server includes a processor, a voltage regulator, and a voltage regulator controller. The voltage regulator is coupled to the corresponding processor to provide actual power to the corresponding processor. The voltage regulator controller is coupled to the voltage regulator, and is used for adjusting the actual power provided by the voltage regulator. The host manager controls the voltage stabilization controller in the server, and uses the voltage stabilization controller to adjust the actual power provided by the voltage regulator to manage the power of the processor.

The power management method of the present invention is applicable to a host manager. The host manager resides in a power management system that includes a server. The power management method includes the following steps: measuring a plurality of parameters in the server to monitor the server, wherein The server includes a processor, a voltage regulator, and a voltage regulator controller; and controlling the voltage regulator controller, and using the voltage regulator controller to adjust the actual power provided by the voltage regulator to manage the voltage regulator Processor power.

Based on the above, the host manager in the power management system of the embodiment of the present invention uses the voltage stabilization controller in each server to adjust the actual power of each processor, thereby managing the power of the processor, so that the power of the processor can be moderately Exceeding the upper limit of its thermal design power or reducing the power of the processor has reduced the overall power consumption. For example, the actual power of the processor can be reduced by using the over-voltage return method of the voltage stabilization controller, or the actual power of the processor can be appropriately increased by using the low-power return method of the voltage stabilization controller. In this way, the host manager in the power management system can set and adjust the actual power of the internal processors of multiple running servers at the same time, so as to instantly increase or limit the overall power consumption of the server. In addition, the set and running server can reach the power management of this embodiment without power cycling such as restarting, which can save processing time.

In order to make the above features and advantages of the present invention more comprehensible, embodiments are hereinafter described in detail with reference to the accompanying drawings.

100‧‧‧Power Management System

110‧‧‧Host Manager

120-1, 120-2‧‧‧ server

130-1, 132-1, 130-2, 132-2‧‧‧ processors

130-1, 132-1, 130-2, 132-2‧‧‧‧ Regulators

150-1, 150-2‧‧‧ voltage regulator controller

160-1, 160-2‧‧‧ substrate management controller

170-1, 170-2‧‧‧ sensors

S210 ~ S250‧‧‧step

310, 320, 330, 410, 420, 430, 440‧‧‧ lines

FIG. 1 is a block diagram of a power management system according to an embodiment of the present invention.

FIG. 2 is a flowchart of a power management method according to an embodiment of the present invention.

FIG. 3 and FIG. 4 are the performance improvement modes of the server 120-1 in FIG. The servo 120-2 in 1 is a schematic diagram of the total power / working frequency in the power saving mode.

FIG. 1 is a block diagram of a power management system 100 according to an embodiment of the present invention. The power management system 100 includes a host manager 110 and at least one server (here, servers 120-1 and 120-2 are taken as examples). The power management system 100 of this embodiment is applicable to a data center with a large number of servers. The host manager 110 may be a specific host server used to manage these servers, or one of these servers.

The internal structure of the server 120-1 is described here. The server 120-1 mainly includes at least one processor (e.g., processors 130-1 and 132-1), and a voltage regulator (e.g., voltage regulators 140-1, 142-1) corresponding to each processor And the voltage regulator controller 150-1. The processor in this embodiment uses a central processing unit (CPU) as an example. Those who apply this embodiment can adjust the type of processor according to their needs. For example, a graphics processor (GPU) or other type of microprocessor can be used. As the processor according to the embodiment of the present invention. The voltage regulators 140-1 and 142-1 are respectively coupled to the corresponding processors 130-1 and 132-1. The voltage regulators 140-1 and 142-1 provide actual power to the corresponding processors 140-1 and 142-1, respectively. The voltage regulator controller 150-1 is coupled to each voltage regulator (for example, voltage regulators 14-10 and 142-1 in the server 120-1), and is used to adjust the voltage regulators 140-1, 142- The actual power provided to each processor 130-1, 132-1.

The server 120-1 further includes a substrate management controller 160-1. The baseboard management controller 160-1 in the server 120-1 can be managed by the host through the network or other methods. 器 110 通信。 110 communication. The server 120-1 also includes various types of sensors 170-1, which are coupled to the substrate management controller 160-1. The sensor 170 is used to measure a plurality of parameters in the server 120-1 (such as temperature, power consumption, operating voltage, operating current, etc. of each processor). The baseboard management controller 160-1 transmits the parameters measured by the sensor 170-1 to the host manager 110, so that the host manager 110 monitors the server 120-1.

The server 120-2 has an internal structure similar to that of the server 120-1. That is, the server 120-2 includes processors 130-2, 132-2, voltage regulators 140-2, 142-2, voltage regulator controller 150-2, substrate management controller 160-2, and sensor 170. -2.

The embodiments of the present invention do not limit the number of processors, voltage regulators and voltage regulator controllers corresponding to the processors in a single server. For example, those who apply this embodiment may have only a single processor and a corresponding single voltage regulator in the server according to their needs, and may also set two, four, or even 16 processors in the server and a corresponding number of stable voltage regulators. Press. The number of voltage regulator controllers will be adjusted accordingly due to the number of voltage regulators it can control simultaneously.

Generally speaking, the voltage stabilization controller 150-1 located in the server 120-1 is to allow each processor 130-1, 132-1 to obtain an appropriate power, and the value of this power should not exceed each processing as much as possible. The upper limit of the rated power (that is, the thermal design power (TDP)) set at the time of leaving the factory 130-1, 132-1. It is assumed here that the upper limit of the rated power of the processors 130-1 and 132-1 is 100W. However, due to the materials, power supply load, etc. of the regulators 140-1, 142-1, etc., it may cause the regulators 140-1, 142-1 to output power that exceeds 100W (for example, 105W) for processing.器 130-1, 132-1. At this time, the voltage regulator controller 150-1 needs to adjust the outputs of the voltage regulators 140-1 and 142-1 and reduce them. Low is 100W to meet the needs of processors 130-1, 132-1. The processors 130-1 and 132-1 adjust the operating frequency and power consumption according to their own workloads, so as to perform their own power management, but the processors 130-1 and 132-1 cannot increase their upper power consumption limits.

From the above, it can be seen that the voltage stabilization controller 150-1 can actually increase or decrease the output power of the voltage regulators 140-1 and 142-1 moderately, that is, the actual power consumption obtained by each processor can be Not equal to the rated power consumption set by the processor at the factory. Therefore, the embodiment of the present invention facilitates using the "stabilizing controller 150-1 to adjust the output power of the regulators 140-1 and 142-1" as the power management system 100 for each processor (for example, the processor 130-1 , 132-1, 130-2, 132-2). In this way, although each processor has its own rated power upper limit, this embodiment can moderately increase the actual power of each processor to be greater than its rated power upper limit, so that it temporarily obtains higher working efficiency, so that To cope with the temporary need for the server to obtain a higher workload, reduce the possibility of overloading the server with high power consumption (for example, game manufacturers will flood a lot of people when the game is just started, or during a certain period of time Expectedly, a large number of users will come in; when popular tickets are opened for online ordering, the server workload will be expected to increase ... etc.).

In this embodiment, the maximum value of the actual power may be 105% of the upper limit of the rated power of each processor; in other words, this embodiment may raise the upper limit of the rated power by 5% to adjust the Actual power, and set the recovery time in the control conditions. The actual power of the processor has been adjusted back to the original rated power upper limit to prevent the processor from being damaged due to excessive real power and long operating time. In contrast, this embodiment does not limit the range in which the actual power of each processor is adjusted downward, because when When the actual power of the processor is lower than the upper limit of the rated power, the processor is less likely to be damaged.

On the other hand, this embodiment can also reduce the actual power of each processor to be less than its rated power limit, thereby reducing power consumption (for example, limiting the power consumption of the server during high power bill periods; temporarily cutting off the main power supply or short-term When a backup power source is used as a power source when there is a power shortage, the power consumption of the server is temporarily limited to extend the backup power source ... etc.).

Therefore, the host manager 110 of this embodiment may determine whether the server needs to be set to the performance improvement mode (that is, increase the actual power of the processor to be greater than the upper limit of the rated power) according to multiple control conditions set by the user or Power saving mode (ie, reducing the actual power of the processor to less than the upper power rating). The control conditions in this embodiment and the server changes corresponding to these control conditions can be set by the user or manager directly according to their needs. In this way, when a special situation occurs (such as judging that a control condition has occurred or triggered), the manager can directly use the server's predefined control change information corresponding to the control condition that has occurred or triggered. To adjust the servers 120-1, 120-2. In detail, the manager can set a plurality of control conditions (for example, condition 1, condition 2 and condition 3) on the host manager 110 and the predefined server change control information (change 1) that they want to perform when these conditions occur , Change 2 and change 3). When one of the conditions 1 to 3 occurs, the host manager 110 notifies the manager when the occurred condition (eg, the condition 1) occurs, so that the manager has the ability and power to change the control information from a predefined server ( For example, change 1, change 2 and change 3) Select one of them to make the server 120-1, 120-2 make corresponding changes. The host manager 110 may also directly select the corresponding change 1, change 2, and change 3 to control the servers 120-1, 120-2 when one of the conditions 1, 2, and 3 occurs. On the other hand, the host manager 110 can also learn each of the servers 120-1 and 120 according to the parameters of the corresponding servers 120-1 and 120-2 transmitted by the baseboard management controllers 160-1 and 160-2. -2 conditions (such as workload, power supply situation, etc.), so as to determine whether the server is set to the performance improvement mode or the power saving mode, so that the host manager 110 performs the server 120-1, 120- 2 for automated power management.

For example, the user may preset a plurality of first control conditions in the host manager 110. When the host manager 110 determines that one of these first control conditions occurs or is triggered, the server 120-1, 120- 2 Set to performance boost mode. The user may also set a plurality of second control conditions in the host manager 110 in advance. When the host manager 110 determines that one of these second control conditions occurs or is triggered, the server 120-1, 120-2 Set to power saving mode. Even when the host manager 110 determines that the triggered first control condition or the triggered second control condition has been completed or eliminated, the host manager 110 can also control the voltage stabilization controllers 160-1, 160-2 so that Each voltage regulator 140-1, 142-1, 140-2, 142-2 provides rated power to each process 130-1, 132-1, 130-2, 132-2. In addition, the servers 120-1 and 120-2 that are set and running can achieve power management in this embodiment without using a power cycle such as restarting, which can save processing time.

FIG. 2 is a flowchart of a power management method according to an embodiment of the present invention. Illustration. This power management method is applicable to the host manager 110 in FIG. 1, and the host manager 110 is located in the power management system 100 including the servers 120-1 and 120-2. Please refer to FIG. 1 and FIG. 2 at the same time. In step S210, the host manager 110 uses the sensors 170-1 and 170-2 on the servers 120-1 and 120-2 to measure the servers 120-1 and 120. -2 multiple parameters to monitor the servers 120-1, 120-2.

In step S220, the host manager 110 determines whether a plurality of first control conditions and a plurality of second control conditions occur or are triggered. As can be seen from the above description, a user or a maintenance person can set a plurality of first control conditions and a plurality of second control conditions in the host manager 110 in advance.

When one of the plurality of first control conditions and the plurality of second control conditions has occurred or triggered, the process proceeds from step S220 to step S230, and the host manager 110 controls the corresponding stability in the 120-1 and 120-2 servers. Voltage controller 150-1, 150-2, and use the voltage regulator controllers 150-1, 150-2 to adjust the actual power provided by the regulators 140-1, 142-1, 140-2, 142-2 to manage Power supply for processors 130-1, 132-1, 130-2, 132-2. In detail, the host manager 110 learns that the server is in the performance improvement mode or the power saving mode according to the type of the control condition (the first control line or the second control condition), and transmits the control commands of the corresponding mode to the substrate respectively. Management controllers 160-1, 160-2. The baseboard management controllers 160-1 and 160-2 adjust the actual power provided by the regulators 140-1, 142-1, 140-2, and 142-2, respectively, according to the control instructions.

Step S230 can be divided into multiple detailed steps S232, S234, and S236, which are described here one by one. In step S232, the host manager 110 uses a preset Multiple first control conditions and multiple second control conditions to determine whether the servers 120-1, 120-2 are in the performance improvement mode or the power saving mode. It is assumed here that the server 120-1 is judged as a performance improvement mode, and the server 120-1 is judged as a power saving mode.

When the host manager 110 determines that the server 120-1 is in the performance improvement mode, the process proceeds from step S232 to step S234, and the host manager 110 controls the voltage stabilization controller 160-1 so that it uses an under reporting method to allow The regulators 140-1, 142-1 provide actual power (eg, 135W) higher than the rated upper power limit (eg, 120W) of the corresponding processors 130-1, 132-1 to the processors 130-1, 132- 1. Therefore, the working frequencies of the processors 130-1 and 132-1 can be higher than the original preset performance. Although providing the actual power higher than the rated power upper limit to the processors 130-1 and 132-1 will lead to an increase in power consumption, it also allows the processors 130-1 and 132-1 to have higher performance. There is no need to add additional redundant servers.

On the other hand, when the host manager 110 determines that the server 120-1 is in the power saving mode, the process proceeds from step S232 to step S236, and the host manager 110 controls the voltage stabilization controller 160-2 to make it use the power-return (power- over reporting) to enable the regulators 140-2, 142-2 to provide the actual power (e.g., 110W) below the rated power upper limit (e.g., 120W) of the corresponding processors 130-2, 132-2 to the processor 130 -2, 132-2. This can achieve the effect of saving power consumption. In other words, the host manager 110 directly reduces the actual power of the processors 130-2 and 132-2 of the server 120-2 through the voltage stabilization controller 160-2, so that the overall power consumption of the server 120-2 Adjust directly down.

In step S240, the servers 120-1 and 120-2 are in the performance improvement mode. In the power saving mode, the host manager 110 determines whether the triggered first control condition or the triggered second control condition has been completed or eliminated. When the triggered first control condition or the triggered second control condition is not completed or eliminated, the host manager 110 will continuously maintain the servers 120-1, 120-2 in the performance improvement mode or the power saving mode In step S240, the determination is continuously performed. In contrast, when the triggered first control condition or the triggered second control condition is completed or eliminated, the process proceeds from step S240 to step S250, and the host manager 110 controls the voltage stabilization controllers 160-1, 160-2. In order for the corresponding voltage regulators 140-1, 142-1, 140-2, 142-2 to provide the upper rated power (ie, 120W) to the processors 130-1, 132-1, 130-2, 132-2 .

3 and 4 are schematic diagrams of the total power and operating frequency when the server 120-1 is in the performance improvement mode and the server 120-2 is in the power saving mode, respectively. The horizontal axis of FIG. 3 represents time, and the vertical axis of FIG. 3 represents power. As shown in FIG. 3, line 310 represents the power consumption measured by the processor 130-1 in the server 120-1, and line 310 also represents the power consumption measured by the processor 130-2 in the server 120-2. It can be known from FIG. 3 that the lines 310 are all 125W. Here, the total power is measured from the outside of the server 120-1 and the server 120-2, so that it is known that the line 320 is the total power consumption of the server 120-1 in the performance improvement mode, and the line 330 is the power saving mode The total power consumption of the server 120-2. It can be easily seen from FIG. 3 that the total power consumption of the server 120-1 in the performance improvement mode is greater than the total power consumption of the server 120-2 in the power saving mode.

The horizontal axis of FIG. 4 represents time, and the vertical axis of FIG. 4 represents the operating frequency of the processor. As shown in FIG. 4, line 410 represents the position of the server 120-1 in the performance improvement mode. The operating frequency of the processor 130-1; line 420 represents the operating frequency of the processor 132-1 of the server 120-1 in the performance improvement mode; the line 430 represents the operating frequency of the processor 130-2 of the server 120-2 in the power saving mode Operating frequency; line 440 represents the operating frequency of the processor 132-2 of the server 120-2 in the power saving mode. As can be seen from Figures 3 and 4, although the processors 130-1 and 132-1 of the server 120-1 have higher power consumption, they also have higher operating frequencies to carry higher workloads; The processors 130-2 and 132-2 of the processor 120-2 have lower power consumption and lower operating frequencies to save power.

In summary, the host manager in the power management system of the embodiment of the present invention uses the voltage stabilization controller in each server to adjust the actual power of each processor, thereby managing the power of the processor, so that the power of the processor can Moderately exceeding the upper limit of its thermal design power or reducing the power of the processor has reduced the overall power consumption. For example, the actual power of the processor can be reduced by using the over-voltage return method of the voltage stabilization controller, or the actual power of the processor can be appropriately increased by using the low-power return method of the voltage stabilization controller. In this way, the host manager in the power management system can set and adjust the actual power of the internal processors of multiple running servers at the same time, so as to instantly increase or limit the overall power consumption of the server. In addition, the set and running server can reach the power management of this embodiment without power cycling such as restarting, which can save processing time.

Although the present invention has been disclosed as above with the examples, it is not intended to limit the present invention. Any person with ordinary knowledge in the technical field can make some modifications and retouching without departing from the spirit and scope of the present invention. The protection scope of the present invention shall be determined by the scope of the attached patent application.

Claims (11)

A power management system includes: a host manager; and a server in communication with the host manager, wherein the server includes: a processor; a voltage regulator coupled to a corresponding processor to provide actual power To the corresponding processor; and a voltage regulator controller coupled to the voltage regulator to adjust the actual power provided by the voltage regulator, wherein the host manager controls the servo The voltage regulator controller in the controller, and using the voltage regulator controller to adjust the actual power provided by the voltage regulator to manage the power of the processor. The power management system according to item 1 of the scope of patent application, wherein the server further comprises: a baseboard management controller, which communicates with the host manager through a network, wherein the host manager transmits a control instruction to the host The baseboard management controller adjusts the actual power provided by the voltage regulator according to the control instruction. The power management system according to item 2 of the scope of patent application, wherein the server further comprises: a plurality of sensors coupled to the substrate management controller, and the sensors measure the server A plurality of parameters, wherein the substrate management controller transmits the parameters measured by the sensor to the host manager, so that the host manager monitors the server. The power management system according to item 1 of the scope of patent application, wherein the host manager sets a plurality of first control conditions in advance to determine whether the server is in a performance improvement mode, and responds to the host manager to determine the When the server is in the performance improvement mode, the host manager controls the voltage regulator controller so that it uses an under reporting method to allow the voltage regulator to provide higher than the corresponding processor The rated power is capped by the actual power to the processor. The power management system according to item 4 of the scope of patent application, wherein the maximum value of the actual power is 105% of the upper limit of the rated power of the processor. The power management system according to item 4 of the scope of patent application, wherein the host manager further presets a plurality of second control conditions to determine whether the server is in a power saving mode, and responds to the host manager to determine When the server is in the power saving mode, the host manager controls the voltage regulator controller so that it uses a power-over reporting method so that the voltage regulator provides The actual power of the rated power upper limit of the processor is given to the processor. The power management system according to item 6 of the scope of patent application, wherein when the server is in the performance improvement mode or the power saving mode, the host manager determines that the first control condition or Whether the triggered second control condition has been completed or eliminated, and in response to the host manager judging that the triggered first control condition or the triggered second control condition has been completed or eliminated, The host manager controls the voltage regulator controller to allow the voltage regulator to provide the rated power cap to the processor. A power management method is applicable to a host manager, the host manager is located in a power management system including a server, and the power management method includes: measuring a plurality of parameters in the server to monitor the server And the server includes a processor, a voltage regulator, and a voltage regulator controller; and controls the voltage regulator controller, and uses the voltage regulator controller to adjust the actual power provided by the voltage regulator to manage A power source for the processor. The power management method according to item 9 of the scope of patent application, further comprising: setting a plurality of first control conditions in advance to determine whether the server is a performance improvement mode; and determining the server as the performance in response In the boost mode, the voltage stabilization controller is controlled so that it uses the low power return mode to allow the voltage regulator to provide the actual power to the processor that is higher than the corresponding rated power upper limit of the processor. The power management method according to item 9 of the scope of patent application, further comprising: presetting a plurality of second control conditions to determine whether the server is in a power saving mode; and determining the server as the power in response In the saving mode, the voltage stabilization controller is controlled to make the voltage regulator to provide the actual power below the rated power upper limit of the corresponding processor to the processing by using an over-return mode. Device. The power management method according to item 10 of the scope of patent application, further comprising: when the server is in the performance improvement mode or the power saving mode, the first control condition or the triggered Whether the second control condition has been completed or eliminated; and in response to the host manager determining that the first control condition that has been triggered or the second control condition that has been triggered has been completed or eliminated, controlling the The voltage regulator controller causes the voltage regulator to provide the rated power cap to the processor.